Adding new node with old node's name

[Mike Smith_33]

Before I do this I want to run it past the group in case I am missing something. I had a quorum node die on me. We have a spare node to replace it with. Normally you would look to do a recovery from the dead system's boot disk and get up pretty quickly. In this case, they decided to go with a portable shared tape drive and only do system disk backups in leap years. Now that the node is dead, the tape drive is nowhere to be found.

I figured I probably have at least a couple of options. Try to pass off the new guy as the old and just install VMS and add him to the cluster. Another option could be removing the old node from the cluster and then adding this as a totatlly new member.

I have started down the path of passing this guy off as the old node. VMS 7.3-2 has been installed and it is ready for addition to the cluster.

Any concerns? The other 2 nodes are in production. I think the chance of crashing the cluster is remote but still, there is a chance.

Should I scrap this and go the other way or would a 3rd or 4th option be better? Any thoughts??

All nodes run 7.3-2, clustered over the network, individual system disks in each with shared drive location for cluster common files.
This node is just for quorum, cluster is 3 node cluster, currently only 2 until this one is added.

[Andy Bustamante · Accepted solution]

You can easily do this. You'll want to make sure the prior node name matches the prior DECnet address, at least in phase IV. If you try to use the same name with a different address the new node won't be allowed to join the cluster.

I've migrated a cluster from VAX-4000 to 7000s to Alpha 4100s, ES-40, GS-80s and GS-1280 all with same node names and DECnet ids.

Make sure to set the cluster id and password, or copy CLUSTER_AUTHORIZE.DAT from the present cluster.

Andy

[Mike Smith_33]

FYI: I will wait until I get a sample of answers before handing out points.


Andy, thanks for the response. I set the scsnode id the same. We don't run any decnet so that one was easy. I am trying my best to dress this guy up like the old node so that the cluster really doesn't notice.

[Jan van den Ende]

Mike,

you want some more "flesh on the bones". Ok.

I totally concur with Andy. Nodename and SCSSYSTEMID should BOTH be exactly as the previous member, OR, BOTH different.

Copying CLUSTER_AUTHORIZE.DAT is also a good suggestion, much better than trying to (remember and) typing error-free.

And yes, I have also done such exercises a few times. Pretty smooth going; after all, this IS VMS!

Success!

Proost.

Have one on me.

jpe

[Robert Gezelter]

Mike,

I concur.

All the cluster should be aware of is the SCSSYSTEMID and the Cluster Group number and passord.

- Bob Gezelter, http://www.rlgsc.com

[comarow]

To throw in a monkey wrench, if the node
every boots with ONE either SCSNODE or
SYSSTEMID correct and one wrong, you will
never be able to bring this node into the
cluster, even if corrected, until you reboot
the entire cluster.

It protects you.

[Volker Halle]

re. Bob,


All the cluster should be aware of is the SCSSYSTEMID and the Cluster Group number and passord.


Both SCSSYSTEMID AND SCSNODE need to match the old node (if the old node is still 'known' by the nodes in the cluster - SHOW CLUSTER will tell) or you must use different values for both parameters. If the cluster group number or password do not match the existing cluster, your node will not be able to join the existing cluster. If the group number is wrong, the existing cluster doesn't even notice your new node. If just the password is wrong, you'll get lots of errors on PEA0 (and in ERRLOG.SYS).

Make sure you set the correct values for VOTES and - especially - EXPECTED_VOTES.

Volker.

[Mike Smith_33]

Ok, I finally got the node added in today. Getting a copy of the existing cluster_authorization.dat over to the new node and also a copy of lmf. Basically I installed vms, got the node up enough that I could load the tcpip license. Opened up ftp and moved necessary files over. I ran into a problem because scssystemid was set incorrectly (I had to guess and I got it narrowed to two choices, of course I chose the wrong one) I got the scssystemid corrected today and the node came right into the cluster.

Thanks for all the responses.